goern-skills

A collection of Claude Code skills by goern.

Skills

deep-research

Turns a single research prompt into a well-cited markdown report and a companion blog post through parallel multi-agent research, iterative synthesis, and citation verification.

Inspired by Anthropic's multi-agent research system.

How It Works

Prompt → Plan → Parallel Research (2-8 subagents) → Synthesis → Gap-Fill → Citation Verification → Report → Blog Post

1. Plan — classifies complexity, generates MECE research axes
2. Research — spawns parallel subagents, each searching and fetching independently
3. Synthesize — merges findings, deduplicates, identifies gaps and contradictions
4. Gap-Fill — spawns targeted subagents for missing coverage (max 3 iterations)
5. Verify — dedicated CitationAgent checks every claim against its source URL
6. Write — produces a single markdown file with verification badges
7. Blog — generates a 1-2 minute companion blog post summarizing key findings in first-person perspective

Each report uses the Outcomes / Outputs / Results framework and includes a full bibliography with credibility ratings and verification status.

Installation

claude plugin marketplaces add https://github.com/goern/goern-skills
claude plugin install goern-skills

Or for development:

claude --plugin-dir /path/to/goern-skills

Usage

/goern-skills:deep-research "What is RLHF?"
/goern-skills:deep-research "Compare RAG vs fine-tuning for enterprise LLM deployment"
/goern-skills:deep-research https://arxiv.org/abs/2401.12345

Output is written to ./research-output/:

• Full report: <slug>-<date>.md
• Blog post: <slug>-blog-<date>.md

Complexity Scaling

Complexity	Subagents	When
Simple	1–2	Single concept, definitional
Moderate	3–4	Comparison, multi-faceted, emerging field
Complex	5–8	Cross-domain, controversial, primary sources needed

Output Sections

• Executive Summary
• Key Findings (with inline citations and verification badges)
• Analysis
• Outcomes / Outputs / Results (hypothesis chain)
• Quotations (verbatim, with verification status)
• Sources / Bibliography (with credibility ratings)
• Methodology (subagents, iterations, gaps)
• Verification Summary (verified / partial / unverified / unavailable)

Project Structure

.claude-plugin/
  plugin.json             # Plugin manifest
skills/
  deep-research/
    SKILL.md              # 7-phase orchestration prompt
    references/
      output-template.md  # Report template
      blog-post-template.md # Blog post template (1-2 min read, first-person)
      research-guide.md   # Complexity heuristics, source quality, search strategy

License

https://spdx.org/licenses/GPL-3.0-or-later.html